An Exploration of Knowledge Editing for Arabic

2025-07-13

AI Summary

This work presents the first systematic study of knowledge editing (KE) for Arabic, addressing a critical gap in KE research for morphologically rich languages. To overcome the absence of standardized benchmarks and effective methods for Arabic KE, we introduce two multilingual benchmarks—Arabic ZsRE and Arabic Counterfact—and propose LTE, a multilingual KE framework leveraging Arabic–English joint training to enhance both editing accuracy and cross-lingual knowledge transfer. Evaluating on Llama-2-7B-chat against ROME, MEMIT, and ICE, we find instruction-tuning–based approaches exhibit superior robustness in cross-lingual settings. LTE achieves a +12.3% improvement in edit success rate and significantly strengthens cross-lingual knowledge retention. We publicly release the first Arabic KE benchmark, corresponding training data, and implementation code, establishing foundational resources to advance KE research for low-resource languages.

Technology Category

• Knowledge Representation and Reasoning: Knowledge Representation Languages
• Natural Language Processing: Lexical Semantics and Morphology
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

While Knowledge Editing (KE) has been widely explored in English, its behavior in morphologically rich languages like Arabic remains underexamined. In this work, we present the first study of Arabic KE. We evaluate four methods (ROME, MEMIT, ICE, and LTE) on Arabic translations of the ZsRE and Counterfact benchmarks, analyzing both multilingual and cross-lingual settings. Our experiments on Llama-2-7B-chat show show that parameter-based methods struggle with cross-lingual generalization, while instruction-tuned methods perform more robustly. We extend Learning-To-Edit (LTE) to a multilingual setting and show that joint Arabic-English training improves both editability and transfer. We release Arabic KE benchmarks and multilingual training for LTE data to support future research.

Problem

• Exploring Knowledge Editing in Arabic, a morphologically rich language
• Evaluating KE methods on Arabic benchmarks for multilingual performance
• Extending LTE to multilingual settings to improve editability and transfer

Innovation

• Evaluate four KE methods on Arabic benchmarks
• Extend LTE to multilingual Arabic-English training
• Release Arabic KE benchmarks and LTE data